Workflows Speed Optimization
First of all, This is not a criticism but a suggestion.
I'm a BIGG FAN of your workflows, I don't use anything but your workflows, nothing else and that's how I discovered this.
I was using using your LTX-2.3_-_I2V_T2V_Basic workflow as a test for LTX 2.5 just to see if I could use your workflows with new from now on and yeah in fact it worked as expected. Then I thought "What's is new on the official comfyui templates" and I saw there's absolutely nothing new, BUT the comfyui template generation was 2x faster than yours, and I couldn't understand why. I took a look at the subgraph and didn't find any clue.
So I thought I'd should tell you this. So my suggestion is if you could take a look at this and find what is slowing down your workflow it would be great since this 2x speed gain and this would mean a lot with complex workflows like a the ID-Lora workflow which I use a lot and it's significantely slower than the basic I2V workflow.
Again - Thank you very much for all your work.
Please see the screenshot below. The longer times are with your workflow and the shortest one is with comfyui's template
Dont see any reason for that, that is obvious at least ; -)
But will take a look
There is not really anything different about my wf and the stock default wf's, at least in theory. All they are is organized a bit differently (for the basic t2v and i2v).
(there is a lot of other wf's here that the stock wf's do not have, but the basic i2v and t2v should behave same)
In your screenshot it could be as simple as the vae decoder. Tiled vs not tiled
I see you had an out of memory. So that slowed down everything, and switched from regular to tiled.
That was probably the reason for the slow down ; ) Think most of my WFs have both the fast regular vae decode and the tiled vae decode that you can swap out.
Its tricky to be generic about it, since if your specs are good enough regular vae will be fast, but if you are on lower specs a tiled vae might be needed
(the regular one will auto switch to tiled, but does make a little slow down since it tries both)